LM2596: Inverting mode instability

Hi Paul,

Could you try to change the inductor on the - 12V output into 100uH and see if the oscillation stops?

Thanks

-Arief

Hi Arief,

Thanks for replying!  I actually started out with an 100uH inductor and 68uF cap.  These values were based on the graphs for component selection for positive output supply design.  It was only later that I spotted the section on page 14 of the datasheet about inverting supply.

The oscillation was actually much worse with the 100uH inductor and 68uF cap!  The pic in the post above was for that combination.  I've just quickly taken another reading for the 33uH/220uF combination and added it below:

So should I try a smaller inductor or larger cap?  I'm a little unsure of the trade-offs with altering these components.  I'm assuming there's minimum and maximum recommended values for them.

Any help would be much appreciated!

Hi Paul,

Can we try to modify the -12V output to a -5V output and follow the datasheet component on figure 18?
https://www.ti.com/lit/ds/symlink/lm2596.pdf?ts=1594248124027&ref_url=https%253A%252F%252Fwww.google.com%252F

This way we can eliminate if there is any other issue unrelated to the component selection. 

A smaller Cout value might help since it will push the LC resonance to a higher frequency.

Have you try the combination of 33uH with 68uF capacitor?

Thanks

-Arief

Thanks

-Arief

Hi Arief,

I tried switching the cap to 68uF, but not much difference in the amount of modulation. The resonant frequency did shift a little though.

r.e. testing with -5V design: I'll have to have a think about the circuits that create the load.  They contain ICs expecting +/-12V rails, so need to check if it is OK with a -5V rail.  It also uses a +24V input supply and fig 18 in the datasheet uses a 4.5-20V input supply.  I might be able to replace the load circuits with a high power resistor, as the PSU and load circuits are on separate PCBs.

Hi Paul,

Could you add a ceramic capacitor between pin 1 and 3 of the device to minimize current loop?

Also how did the routing between the pin 4 to sense the output voltage? Is this a 4 layer board or 2 layer board? I wonder if the sensing trace is right underneath the SW node. 

Here is an information on what is the hot loop that may cause some noise issue

https://www.ti.com/lit/an/slva219a/slva219a.pdf?ts=1594477351732&ref_url=https%253A%252F%252Fwww.google.com%252F

Thanks

-Arief

Hi Arief,

Sorry for the delay. I had run out of 470nF ceramic caps!  I've now got some and have tried adding one where you suggested. There's a slight difference in the resonant frequency, but not a huge change.

Regarding the SW trace and pin 4 trace, these do in fact run under each other as you mention.  The board is only 2 layer though, not 4 layer.  Here's the 2 traces:

So could this layout be the issue?  I am very tight for space, so couldn't follow the datasheet guide. Also, the TO220 package must be perpendicular with the top edge of the PCB, not 90° (as in the datasheet).

But I could do a redesign if it eliminates the noise. I'll just have to allow more space for it!

One question:  what is the maximum specified current for the LM2596 in inverting mode?  I presume it is less than 3A, but surely more than the 700mA that I require?

many thanks for you help!

Hi Paul,

The maximum current for LM2596 in inverting mode is roughly around 2A based on your 24VIN and -12VOUT and it is dependent upon your duty cycle. The equation is Iout < (1-D) * 3A as found in this app note. https://www.ti.com/lit/an/snva722b/snva722b.pdf

The layout is a little tricky in the inverting buck boost connection. There are 2 loops that you have to worry about and make small. 
The first loop is from the input capacitor through the internal high side FET and inductor 
The second loop is from the output capacitor through the diode and inductor. 

I am trying to achieve a small loop like mentioned in this article:  https://e2e.ti.com/blogs_/b/powerhouse/archive/2016/09/28/laying-out-an-inverting-buck-boost-converter-for-success

You can also try to move Z3 to be across pin 2 and 3 in addition to adding ceramic capacitor between pin 1 and 3. This is to minimize the loop and see if it makes any difference. 

Also we can try to add capacitor between the input path to actual ground, since C22 is referred to the VOUT due to inverting buck boost configuration. 

It is based on this app note. https://www.ti.com/lit/an/snva856a/snva856a.pdf?ts=1594915721191&ref_url=https%253A%252F%252Fwww.google.com%252F

Is your jitter got worse with increasing current? Another alternative to check your component value is that we can send you an EVM and it can be modified to inverting buck boost configuration by putting the input supply between VIN and VOUT and the GND for the EVM becomes your new negative VOUT. This way you may be able to at least test your component selection value. 

Thanks

-Arief

Hi Arief,

Thanks for all the info.  I'll go through it all.  I suspect that it'll be necessary for me to redesign this board and make the trace layout more similar to the EVM board.  I'll move Z3 and the caps closer to the 2596 too.

Regarding the current. I have just tested some different loads.  As I decrease the load, there is a very bad oscillation that I hadn't noticed before.  The pulse width narrows too.  Does this provide any clues to the issue?

Hi Paul,

The oscillation in there is normal and it is due to the inductor current reaches 0A and the inductor rings with the diode parasitic capacitance. 

We call this ring is a DCM ring (Discontinuous Conduction Mode) and this is because you are at light load condition. 

One of my colleague mentioned that adding diode D1 like in figure 20 of the datasheet may help with stability.

This is more of a passed down "tale" since all the designers for this part have no longer with the company. But it may worth a try to see if it helps. 

Otherwise you may be able to use a newer part for your design which have a better simulation model to simulate negative buck configuration such as LMR33630. 

Thanks

-Arief

Hi Arief,

Sorry for the long delaying in replying. I have been completely redesigning and prototyping the board with a component layout that matches the data sheet and demo board as closely as possible.  I've now got big ground planes and short traces where I need them.  Sadly this hasn't helped the issue :(  In fact it may have got worse!!  I haven't yet put a scope on it.  That's just my opinion from listening to the audio output.

However, while I was assembling the new PCB, I spotted something interesting.  Both the -12V and +12V LM2596 have large heatsinks on them because they get very hot. In the final product, the aluminium enclosure will be used to dissipate the heat.  If I put my hand on the heat sink and move it about, the amplitude and frequency of the resonance changes. Is there a possibility that the heatsink acts an aerial somehow? I couldn't find much info in the data sheet to confirm or deny this.  Or any recommend heatsink designs.

My next step is to try using some sheet aluminium instead of a heatsink, to simulate the final enclosure and see how this effects it.

Paul

P.S.  Here's the heatsinks attached to the LM2596s. Note that although they are very close to each other, they are electrically insulated.

Hi Paul,

Yes that seems weird. Since the metal tab is GND, therefore maybe you are adding some capacitance when you touch the heatsink. Is the tab shielded from the heatsink? 

We can send you an EVM if you like to test your design components on the EVM. 

Thanks

-Arief

Ahh - but the tab on the -12V 2596 is connected to -12V, not GND.  The tab on the +12V 2596 is connected to ground though.  The enclosure will be grounded and the -12V 2596 will use a mica insulator kit, to prevent a short between -12V and GND.  My concern is that the close proximity of the +12V and -12V 2596s might be causing the resonance issue.

I'm hoping to have the enclosure in just over a week, so I can test if there is a difference between using the enclosure as a heatsink vs the big heatsinks shown in the photos above.

R.e. the EVM, yes it would be really good to get one.  Then I can see if there's any difference in the amplitude of the resonance and also if the SMD version can dissipate enough heat.  My worry is that I need the Leaded version, so I can connect it to the enclosure to dissipate the heat.  But I'd definitely go with the SMD version if it works OK.  Let me know how I can obtain one.

many thanks
Paul